Wrong efficient signs on regression

[Kezia Akwaboa]

Hi everyone,


I am carrying out a study on the effect of unemployment on happiness(individual satisfaction).
The methodology of previous studies I have been using used ordered probit and ordered logit regression on STATA.
I attempted to do the same for my data set, I am using the UKHLS(UK Household longitudinal study).
Literature shows that there should be an inverse relationship between unemployment and individual satisfaction.
However, when I regress my dep (happiness score) with independent dummy variable unemployment as well as other variables I get the wrong coefficient signs.

Can anyone advise me with what I am doing wrong and why I am getting the wrong coefficient signs?

Much thanks.

[Joseph Coveney]

What's the omitted category for jbstat? It seems that just about anything is preferable to it, including long-term sickness and disability. Or maybe your outcome scale is reversed?

[Kezia Akwaboa]

Sorry, my mistake screenshot omitted two categories:
The omitted is category Self-employment

Using ordered logistic regression with my happiness indicator as my dependent variable(scghq2)
I have attached the description of this variable, I think you have made a valid point here in regard to my outcome scale being reversed. Do you know if I could flip this round?

If anyone could help that would be much appreciated.

[Rich Goldstein]

for useful citations, see #3 in https://www.statalist.org/forums/for...ols-regression

[William Lisowski]

If I understand correctly, your dependent variable scghq2_dv has values from 0 for least distressed to 12 for most distressed. So your dependent variable is a measure of distress (rather than satisfaction), and thus the positive coefficient for "being sick or disabled" indicates that it increases distress, not that it increases satisfaction.

If you want a dependent variable that measures lack of distress (I suppose that means satisfaction) then

generate new_dv = 12 - scghq2_dv

will give you a variable with values from 12 for least distressed (most satisfied) to 0 for most distressed (least satisfied).

[Kezia Akwaboa]

Thank you all much, this has honestly saved my Dissertation. You were certainly right about my outcome variable measuring distress rather than satisfaction. I have changed this now and the coefficients of my variables are much more aligned with what I expected. Have a great day .